Semaglutide developer Novo Nordisk suffers cyberattack; internal pharmaceutical AI models stolen

On June 16, Novo Nordisk A/S—a Danish pharmaceutical company renowned for developing semaglutide, a drug used both for diabetes treatment and weight management—was hit by a cyberattack. Hackers infiltrated Novo Nordisk’s internal infrastructure and exfiltrated massive volumes of data, including pre-trained AI models, proprietary datasets, complete training code repositories, training logs, and container images. Novo Nordisk has confirmed the breach; some clinical trial patient-related data was also compromised. The company has notified Danish regulatory authorities and engaged external cybersecurity experts to investigate. The hackers publicly released an overview of the stolen data: ・ Project 'Dragonfly' AI initiative: 16.7 GB of data—including a fully trained multimodal model supporting text, image, and transcriptomic inputs—ready for immediate inference. ・ Novo Nordisk’s proprietary training dataset: ~407 MB of domain-specific biological and chemical training data, primarily used for model training. ・ Complete source code: ~50 MB, covering model classes, configurations, forward-pass logic, loss computation, and other components of the training pipeline. ・ Full logs from 113 training runs: Used to verify training progress and final outcomes—serving as evidence for the authenticity of the stolen data. ・ Internal infrastructure map: Details of Novo Nordisk’s high-performance computing (HPC) systems, Slurm job scheduler configuration, SSH settings, and related infrastructure. ・ Multiple internal container images: ~53 GB total—containers deployed internally to support specific R&D activities. ・ Other private data: Developer identities and internal hostnames, URLs of private GitHub repositories, CUDA API usage statistics, and kernel/memory operation logs.

China’s first national standard on 'Data Security for Medical AI' is now open for drafting unit and individual participation

Author | Zhihe Standards Center In June 2026, multinational pharmaceutical company Novo Nordisk was attacked by a hacker group, resulting in the theft of 1.3 TB of core data—including information on drugs not yet on the market, clinical trial records, patient privacy data, 30 pre-trained AI models, and 70 datasets. Widely regarded by the industry as a 'commercial espionage incident in the biopharmaceutical sector,' this event sends a clear signal: hackers’ targets have expanded beyond traditional patient data to include AI models themselves. The Novo Nordisk incident is not an isolated case in the pharmaceutical industry but rather a preview of broader data security risks facing medical AI. As healthcare institutions deploy large AI models into mission-critical clinical applications—including decision support, medical imaging analysis, and electronic health record processing—they are generating digital assets comparable to—or even denser than—those held by Novo Nordisk: vast volumes of real-world clinical data, trained model weights, and continuously updated inference outputs. Each constitutes a high-value target for cybercriminals. With attack surfaces expanding from 'data' to 'models,' conventional data security boundaries have been breached. In response, China’s first national group standard for medical AI data security—the 'Specification for Data Security and Privacy Protection in Medical Artificial Intelligence Applications'—has been developed under the management of the China Electronics Chamber of Commerce and organized by the Zhihe Standards Center, with joint drafting by the Information Center of China’s National Medical Products Administration, China Quality Certification Center, and other institutions. This standard establishes a comprehensive, standardized operational framework covering data classification and grading, full-lifecycle security management, privacy-enhancing technologies, and clearly defined responsibilities among multiple stakeholders. Its aim is to resolve the 'data–privacy' dilemma inherent in medical AI and promote standardized, healthy, and sustainable development of the 'AI + Healthcare' industry. 01 Participating Organizations ➣ Managing Organization: China Electronics Chamber of Commerce ➣ Organizing Drafting Entity: Zhihe Standards Center ➣ Joint Drafting Entities (partial list): ・ Information Center of China’s National Medical Products Administration ・ Shanghai Medical Device and Cosmetics Review & Verification Center ・ Information and Application Security Protection Branch, Chinese Health Information and Health Big Data Society ・ China Quality Certification Center Co., Ltd. ・ Beijing Youlian Zhiyi Technology Co., Ltd. ・ Zhengzhou Evan Technology Co., Ltd. ・ Beijing Data Pioneer Zone Service Co., Ltd. ・ Beijing Xiaoyao Mansion Intelligent Technology Co., Ltd. ・ Beijing Zhi Jingling Technology Co., Ltd. ・ Hangzhou Puer Medical Technology Co., Ltd. ・ Shanghai Heiyan Medical Technology Co., Ltd. ・ Shanghai Sailu Life Sciences Co., Ltd. ・ Shanghai Yibao Technology Co., Ltd. ・ Shenzhen Huatongwei International Inspection Co., Ltd. ・ Tianjin iHealth Medical Equipment Co., Ltd.

Novo Nordisk Flags Patient Data Breach from Some Clinical Trials in Cyberattack

Jun 11, 2026

Novo Nordisk confirmed a cyberattack where unauthorized copies of patient data from certain clinical trials were taken. While core operations remain unaffected, the company has launched an investigation and advised patients to report any unusual occurrences.

reuters

2026-02-25

Novo Nordisk Taps Start-Up for New Oral Weight-Loss Drugs

2026-04-13

Wegovy-maker Novo Nordisk partners with OpenAI to speed drug development

2026-04-23

Novo Nordisk's Oral GLP-1 Drug Meets Primary Goal in Late-Stage Diabetes Study

2026-05-05

Novo Nordisk's first-quarter profit beats forecasts

2026-05-26

Global Firms Bring More Work In-House at India Hubs on AI Boost